Those ide to sata adapters are the best upgrade path for these old computers. I doubt 133mbs could ever be realized, it doesn't go that fast on a PIII on a PCI bus. Like I said earlyer were still hard limited by the CPU and the memmory bandwidth on the 486 platform and even a Overclocked 586 can't play quake very well. You can polish a turd but you cant turn lead into gold. Like everything it is what it is.

VLB is capable of those speeds in theory, its a pure 32bit CPU memory bus, up to 40MHz 32bit.

>it doesn't go that fast on a PIII on a PCI bus
>were still hard limited by the CPU

processor has nothing to do with it if we are taking over the bus. In a similar vein you can saturate 133MB/s PCI bus on a 440BX chipset.

>and the memory bandwidth on the 486 platform

Yes. ram write speed would be the limiting factor, up to ~60MB/s on 486 chipsets. Gives me another idea - custom high speed memory controller for 486 platforms 😀 imagine ram running at the speed of cache.

well, if everything is what it is, a huge amount of this forum and hobby would probably cease to exist. things like a VLB SATA card aren't about polishing a turd, they're about getting the absolute most that's theoretically possible out of a certain type of hardware, so are of value to design

in this particular case, if the adapter was built with the aforementioned chips or similar ones, the cost of it would likely be around $100 to end users purchasing one if a quantity of 1,000 were made

given that the VLB caching IDE controllers, good SCSI cards etc go for $100 or more online, such an adapter would likely sell 1,000 units within this hobby readily, so that's some serious potential profit margin to the person who builds one

because of the voltages involved, you can even do ridiculous things like design the adapter to provide both data and power to disk drives attached to it. surface mount a combo SATA/power connector to the board, put the holes in the PCB to mount a 2.5" SSD on it. also that jmicron chip - I think I have a whole pile of worthless Lian Li SATA to PATA 5.25" bay things here somewhere which employ this exact chip, haha. they were meant for plugging in PATA drives into SATA ports.

going back to VLB I/O capacity here - while VL is theoretically capable of around the same bandwidth as PCI, in practice it generally never exceeded around 60-70MB/sec in burst reads in my own experience benchmarking reads from local cache. I think this may simply be the chips of the time period hitting their maximums of what they could output, or the 486 CPU simply not being able to process more raw data

as for PCI, well, I had no issue hitting the bus ceiling on workstation pentium II/III boards with ATTO SCSI cards doing cache reads or RAID 0/1 reads using 15K disks

rasz_pl wrote on 2021-11-08, 08:36:

VLB is capable of those speeds in theory, its a pure 32bit CPU memory bus, up to 40MHz 32bit. […]
Show full quote

VLB is capable of those speeds in theory, its a pure 32bit CPU memory bus, up to 40MHz 32bit.

>it doesn't go that fast on a PIII on a PCI bus
>were still hard limited by the CPU

processor has nothing to do with it if we are taking over the bus. In a similar vein you can saturate 133MB/s PCI bus on a 440BX chipset.

>and the memory bandwidth on the 486 platform

Yes. ram write speed would be the limiting factor, up to ~60MB/s on 486 chipsets. Gives me another idea - custom high speed memory controller for 486 platforms 😀 imagine ram running at the speed of cache.

I have a "high speed x86 / 486" SOC evaluation board that takes SDRAM. Has ISA and PCI as well as SATA. I really need to figure out what to do with the thing as far as a build goes.

I can get approx 8.8MB/s from my AHA-2842VL (when used flash based SCSI2SD device) which is I believe close to limit of narrow SCSI. About the same as PCI-based narrow SCSI adaptors can do.

Of course wide SCSI implementations can do twice as much, but that's just a different technology. Not sure if there were any VL-Bus wide adaptors.

libby wrote on 2021-11-08, 09:03:

going back to VLB I/O capacity here - while VL is theoretically capable of around the same bandwidth as PCI, in practice it generally never exceeded around 60-70MB/sec in burst reads in my own experience benchmarking reads from local cache. I think this may simply be the chips of the time period hitting their maximums of what they could output, or the 486 CPU simply not being able to process more raw data

That sounds like the actual system RAM limiting it, ppl trying to push the limits with high overclocks are managing 90MB/sec but typical was more like 66, probably lower on like a DX-25.